基于时频域特征的缺血性卒中脑自动调节鉴别模型

Diagnostic model of time-frequency analysis-based dynamic cerebral autoregulation for acute stroke patients

目的在急性缺血性卒中患者(acute ischemic stroke,AIS)中,探究时频域特征对脑自动调节功能受损鉴别模型的应用价值。方法纳入AIS患者,收集患者一般资料,使用TCD测量患者动脉血压和脑血流流速,计算临界关闭压(critical closing pressure,CCP)、血管面积阻力指数(resistance area product,RAP)等临床特征和时频域特征。按平均血流指数(mean flow index,Mx)分为脑自动调节功能异常组、正常组。比较引入时频域特征前后,模型的整体准确率(accuracy,Acc)、灵敏度(sensitivity,Sen)、特异度(specificity,Spe)、ROC曲线下面积(area under ROC,AUC)的变化值,分析时频域特征对鉴别脑自动调节功能受损的效果。结果共入组212名患者,异常组78例,正常组134例。二分类logistic回归分析显示,基于临床特征的模型Sen、Spe、Acc、AUC各为91.03%、94.03%、92.92%,0.978;添加时频特征后模型的Sen、Spe、Acc、AUC值各提高了3.87%、1.49%、2.36%、0.6%。Bootstrap结果表明,添加时频特征后的模型在Sen、Spe、Acc、AUC方面均有稳定地提高,各为3.84%(95%CI:3.80%~3.88%)、1.49%(95%CI:1.45%~1.52%)、2.38%(95%CI:2.35%~2.40%)、0.60%(95%CI:0.59%~0.61%)。结论与仅使用CCP、RAP等临床特征建立模型相比,引入时频特征有利于提高AIS患者脑自动调节功能受损的识别能力。

Objective This study aimed to explore the effect of time-frequency features on the identification model of abnormal cerebral autoregulation for acute stroke patients.Methods Acute stroke patients were included and general information of the enrolled patients was collected.Clinical features(such as critical closing pressure and resistance area product)and the time-frequency domain features were calculated with bilateral middle cerebral artery(MCA)blood flow velocity and arterial blood pressure,which simultaneously measured by using TCD.Patients were enrolled and divided into two groups according to the mean flow index(Mx).To analyze the influence of time-frequency domain features on the identification of abnormal cerebral autoregulation for acute stroke patients,the overall accuracy(Acc),sensitivity(Sen)and specificity(Spe)of model whether conclude time-frequency domain features were compared.Results A total of 212 patients were enrolled,concluding 78 cases in abnormal cerebral hemodynamic status and 134 cases in normal cerebral hemodynamic status.The logistic regression analysis showed that the Sen,Spe,Acc and area under ROC(AUC)of the model were 91.03%,94.03%,92.92%and 0.978 with only clinical features and the improvement of the Sen,Spe,Acc and AUC of the model was 3.87%,1.49%,2.36%and 0.6%,after adding time-frequency domain features.Resampling results showed that after adding time-frequency domain features,the model achieved stable improvements in Sen,Spe,Acc and AUC,which were 3.84%(95%CI:3.80~3.88%),1.49%(95%CI:1.45~1.52%),2.38%(95%CI:2.35~2.40%)and 0.60%(95%CI:0.59~0.61%),respectively.Conclusion The time-frequency domain features can improve the accuracy of the identification model for abnormal cerebral autoregulation.